I don't think I'm disagreeing with you really. I'd love to have a standard language for describing probabilistic models, together with some tools to automate deriving a range of different approximate inference algorithms for them. Perhaps also some heuristics on top of this to pick good defaults for the inference method, which I can override. Ideally also a way to implement my own approximations (more realistically, approximations from papers I want to implement) without having to start from scratch.

I'd prefer to decouple those things though, so I can benefit from the first and the second without waiting for a "sufficiently smart compiler" to arrive which can automate the whole shebang well enough to rely on in a fully general programming language setting. That was what I was expressing skepticism about, although I'd be very happy to be proven wrong.